aws

The AWS MCP Server exposes Claude-friendly tools to run AWS CLI commands through the MCP interface. It wraps the AWS CLI itself so Claude can discover available commands on-demand and execute them, returning results into your chat session. Two primary tools are exposed: aws_cli_help, which retrieves documentation for AWS CLI commands, and aws_cli_pipeline, which executes AWS CLI commands and can pipe output through tools like jq or grep for filtering and formatting. This setup relies on your AWS credentials being available to the underlying CLI, and your IAM policy shaping what Claude is allowed to do. The server supports both interactive help queries and scripted command pipelines, enabling Claude to guide you through complex AWS operations step by step or perform rapid, chained actions as part of a single MCP session.

Prerequisites:

• Python 3.8+ (or as required by the aws-mcp-server package)
• AWS CLI installed and configured with credentials (or appropriate environment/role access)
• uv (for running uvx in Python) installed

Installation steps:

1. Install the AWS MCP server package from PyPI: pip install aws-mcp

2. Install uvx if not already installed (used to run the MCP server): pip install uvx

3. Run the MCP server locally (example): uvx aws-mcp

4. Verify that the server is accessible via your MCP client configuration, using the provided command and arguments as shown in the repository README. If you choose to run via Docker or another transport, follow the Docker instructions in the README and ensure AWS credentials are exposed to the container if needed.

Tips and caveats:

• Ensure AWS credentials are accessible to the underlying AWS CLI (environment variables, credentials file, or IAM role).
• The server uses the standard AWS credential chain, so you can specify AWS_PROFILE or other AWS_* variables as needed in your MCP config.
• When using Docker with the AWS MCP Server, bind mount your ~/.aws directory into the container and consider using streamable-http for web-based clients.
• The transport option AWS_MCP_TRANSPORT supports stdio, sse (deprecated), and streamable-http; prefer streamable-http for web clients.
• For best security, run the server in isolated environments (e.g., Docker) and apply least-privilege IAM policies.
• If you encounter timeouts or output size limits, adjust AWS_MCP_TIMEOUT and AWS_MCP_MAX_OUTPUT accordingly.